Moxibustion method and moxibustion robot based on multi-source feature fusion collaborative control

Abstract

This invention relates to a collaborative control method for moxibustion techniques based on multi-source feature fusion and a moxibustion robot, belonging to the field of medical robot control technology. The method includes: acquiring target data of the moxibustion area and extracting target features from the target data to construct a state vector; inputting the state vector into a trained deep reinforcement learning policy network to obtain action commands output by the deep reinforcement learning policy network; and controlling the moxibustion robot to execute corresponding moxibustion techniques based on the action commands. The collaborative control method for moxibustion techniques based on multi-source feature fusion provided by this invention allows the state vector to characterize the subject's thermosensitive physiological response and micro-motion state. Furthermore, by utilizing a deep reinforcement learning model to output action commands for the moxibustion robot, parameterized adjustment of the moxibustion techniques is achieved. This solves the problems of stiff and non-continuous evolution of traditional moxibustion robot techniques, improving the automation level and flexibility of the moxibustion robot.

Description

Technical Field

[0001] This invention relates to the field of medical robot control technology, and in particular to a collaborative control method for moxibustion techniques based on multi-source feature fusion and a moxibustion robot. Background Technology

[0002] Traditional Chinese medicine moxibustion emphasizes "obtaining qi" and "warming and unblocking," and has extremely high requirements for the distance, frequency, and techniques (such as pecking and swirling) of the moxibustion.

[0003] Existing moxibustion robots mostly use teaching and reproduction or simple PID temperature control, which have the following defects: rigid open-loop control, the robotic arm only moves along a fixed trajectory and cannot sense the dynamic changes in the patient's skin temperature distribution; lack of syndrome differentiation, unable to identify the key physiological response indicator of "skin redness"; poor safety, when the patient makes unconscious body movements, the robot cannot follow or avoid them in real time, which can easily lead to burns.

[0004] Existing control methods for moxibustion robots suffer from low automation and rigid, mechanical techniques. Summary of the Invention

[0005] In view of this, it is necessary to provide a collaborative control method for moxibustion techniques based on multi-source feature fusion and a moxibustion robot, so as to improve the automation level and flexibility of the moxibustion robot.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for synergistic control of moxibustion techniques based on multi-source feature fusion, comprising: Acquire target data of the moxibustion area and extract target features from the target data to construct a state vector; the target data includes infrared thermal spectrum data, visible light image data, and contact torque data of the end effector of the moxibustion robot arm; the target features include acupoint center temperature features, skin redness features, human body displacement features, and torque features; The state vector is input into the trained deep reinforcement learning policy network to obtain the action command output by the deep reinforcement learning policy network; the deep reinforcement learning policy network adopts the SAC algorithm; the action command includes the spatial velocity vector of the robotic arm end effector, trajectory superposition parameters, and heating power of the heat source; Based on the action commands, the moxibustion robot is controlled to perform the corresponding moxibustion techniques.

[0007] In one possible implementation, the deep reinforcement learning policy network is trained with the objectives of maximizing cumulative reward and maximizing policy entropy. The reward function of the SAC algorithm is expressed as follows:

[0008] in, Represents the reward function, Indicates real-time temperature. Indicates the target treatment temperature. This indicates the rate of temperature change in the moxibustion area per unit time. A metric for evaluating the smoothness of a robot's motion trajectory. This indicates a positive feedback item for skin flushing. This indicates a penalty for crossing safety boundaries. , , and These represent the weighting coefficients.

[0009] In one possible implementation, controlling the moxibustion robot to perform the corresponding moxibustion technique based on the action command includes: Based on the aforementioned action commands, the moxibustion robot is controlled to perform rotary moxibustion by superimposing a circular motion trajectory in the XY plane onto its basic pose; or, Based on the aforementioned action commands, the moxibustion robot is controlled to perform sparrow-pecking moxibustion by superimposing a reciprocating motion trajectory in the Z-axis direction on the basis of the basic posture.

[0010] In one possible implementation, the state vector is expressed as follows:

[0011] in, Represents the state vector. Indicates the temperature at the center of the acupoint. This represents the temperature field distribution gradient centered on the acupoint. This represents the skin redness component in the Lab color space. This indicates positional deviation caused by slight movements of the human body. Indicates the end contact torque. This indicates the remaining duration of moxibustion treatment.

[0012] One possible implementation also includes: When the temperature at the center of the acupoint exceeds the preset temperature or the positional deviation exceeds the preset displacement, the moxibustion robot is controlled to stop moxibustion and the robotic arm is driven to retract along the positive Z-axis.

[0013] Secondly, the present invention also provides a moxibustion robot, comprising: A six-axis robotic arm body, control unit, and end effector; The end effector is connected to the six-axis robotic arm body via a flange; The end-efficiency integrated actuator integrates an infrared thermal imager, an RGB camera, a six-dimensional torque sensor, and an electrically heated moxibustion head. The control device is used to execute the moxibustion technique collaborative control method based on multi-source feature fusion as described in any of the above implementations.

[0014] In one possible implementation, the infrared thermal imager, the RGB camera, the six-dimensional torque sensor, the electric heating moxibustion head, and the flange are coaxially mounted.

[0015] In one possible implementation, the infrared thermal imager is used to acquire infrared thermal spectrum data of the moxibustion area; The RGB camera is used to acquire visible light image data of the moxibustion area; The six-dimensional torque sensor is used to collect contact torque data.

[0016] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the moxibustion technique collaborative control method based on multi-source feature fusion described in any of the above implementations.

[0017] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, wherein when the program or instruction is executed by a processor, it is able to implement the steps in the moxibustion technique collaborative control method based on multi-source feature fusion described in any of the above implementations.

[0018] The beneficial effects of this invention are as follows: The moxibustion technique collaborative control method and moxibustion robot based on multi-source feature fusion provided by this invention acquire infrared thermal spectrum data, visible light image data and contact torque data of the moxibustion area, and extract the temperature features of the acupoint center, skin redness features, human displacement features and torque features to construct a state vector. Thus, the state vector can be used to characterize the thermal physiological response of the subject's skin and the micro-motion state of the human body. Furthermore, the action commands of the moxibustion robot are output using a deep reinforcement learning model, realizing parameterized adjustment of moxibustion techniques. This solves the problem of stiff and uncontinuous evolution of traditional moxibustion robot techniques, and improves the automation level and flexibility of the moxibustion robot. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic flowchart of an embodiment of the moxibustion technique collaborative control method based on multi-source feature fusion provided by the present invention; Figure 2 A schematic diagram of the deep reinforcement learning control network architecture provided by this invention; Figure 3 This is a schematic diagram illustrating the principle of generating the sparrow-pecking moxibustion trajectory provided by the present invention; Figure 4 This is a schematic diagram illustrating the principle of generating the circular moxibustion trajectory provided by the present invention; Figure 5 This is a schematic diagram of the structure of the moxibustion robot provided by the present invention; Figure 6 A schematic diagram of the structure of the end-effector integrated execution device provided by the present invention; Figure 7 This is a schematic diagram of the overall process of the present invention; Figure 8 A flowchart illustrating the collaborative control method provided by this invention; Figure 9 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0023] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] This invention provides a collaborative control method for moxibustion techniques based on multi-source feature fusion and a moxibustion robot, which will be described below.

[0026] Figure 1 This is a schematic flowchart of an embodiment of the moxibustion technique collaborative control method based on multi-source feature fusion provided by the present invention, as shown below. Figure 1 As shown, the collaborative control method for moxibustion techniques based on multi-source feature fusion includes: S101. Obtain target data of the moxibustion area and extract target features from the target data to construct a state vector; the target data includes infrared thermal spectrum data, visible light image data, and contact torque data of the end of the moxibustion robot's robotic arm; the target features include acupoint center temperature features, skin redness features, human body displacement features, and torque features; S102. Input the state vector into the trained deep reinforcement learning policy network to obtain the action command output by the deep reinforcement learning policy network; the deep reinforcement learning policy network adopts the SAC algorithm; the action command includes the spatial velocity vector of the robotic arm end effector, trajectory superposition parameters, and heat source heating power. S103. Based on the action command, control the moxibustion robot to perform the corresponding moxibustion technique.

[0027] It should be noted that the moxibustion technique collaborative control method based on multi-source feature fusion provided by the present invention can be applied to the moxibustion robot system. By controlling the moxibustion technique of the moxibustion robot, intelligent moxibustion operation on the acupoint area of ​​the subject can be realized.

[0028] In S101, a unified world coordinate system is established, and multi-source data is collected. Infrared thermal spectrum data, visible light image data, and contact torque data of the moxibustion robot's end effector are collected in real time and synchronously.

[0029] The system performs spatiotemporal registration on the collected multi-source data, extracts target features from the target data, and constructs a state vector. The target features include acupoint center temperature features, skin redness features, human body displacement features, and torque features.

[0030] For example, the central temperature of the acupoint, the radial temperature gradient, the color gamut value of the skin rosacea, and the relative displacement vector can be extracted to construct the state vector at the current moment.

[0031] By extracting target features from infrared thermal spectrum data, visible light image data, and contact torque data to construct a state vector, it is possible to deeply characterize the subject's skin's thermosensitive physiological response (including skin temperature changes and skin redness) and the body's micro-motion state.

[0032] In S102, the deep reinforcement learning policy network can use the SAC (Soft Actor-Critic) algorithm, whose objective function includes maximizing the cumulative reward and maximizing the policy entropy.

[0033] The state vector is input into the trained deep reinforcement learning policy network, which outputs continuous action commands based on the maximum entropy principle. The action commands include the spatial velocity vector of the robotic arm end effector, trajectory superposition parameters, and heating power of the heat source.

[0034] In S103, the action instructions output by the deep reinforcement learning strategy network drive the end-effector to execute the corresponding moxibustion technique trajectory.

[0035] For example, the robot can be controlled to perform rotary moxibustion by superimposing a circular motion trajectory in the XY plane on a reference pose, or to perform sparrow-pecking moxibustion by superimposing a reciprocating motion trajectory in the Z-axis direction on a reference pose.

[0036] The robot controller receives motion commands and uses a composite motion control interface to dynamically superimpose periodic geometric trajectories on the basic pose of the end effector, driving the actuator to complete the moxibustion action.

[0037] The parameterized manipulation adjustment achieved by using a deep reinforcement learning model solves the problems of stiff and uncontinuous manipulation techniques in traditional moxibustion robots, and realizes seamless switching and power coordination between techniques.

[0038] In summary, the multi-source feature fusion-based collaborative control method for moxibustion techniques provided in this invention acquires infrared thermal spectrum data, visible light image data, and contact torque data of the moxibustion area, and extracts acupoint center temperature features, skin redness features, human body displacement features, and torque features to construct a state vector. This state vector can be used to characterize the subject's skin's thermosensitive physiological response and the body's micro-motion state. Furthermore, a deep reinforcement learning model is used to output the action commands of the moxibustion robot, realizing parameterized adjustment of moxibustion techniques. This solves the problems of stiff and non-continuous evolution of traditional moxibustion robot techniques, and improves the automation level and flexibility of the moxibustion robot.

[0039] In some embodiments of the present invention, the expression for the state vector is as follows:

[0040] in, Represents the state vector. Indicates the temperature at the center of the acupoint. This represents the temperature field distribution gradient centered on the acupoint. This represents the skin redness component in the Lab color space. This indicates positional deviation caused by slight movements of the human body. Indicates the end contact torque. This indicates the remaining duration of moxibustion treatment.

[0041] The raw data stream containing infrared thermal images, RGB visual information, and torque information collected by the multi-source sensing processing module is acquired. Spatiotemporal alignment and feature extraction are performed on the raw data stream to construct a 32-dimensional state vector.

[0042] The 32-dimensional state vector is input into the pre-trained reinforcement learning model, which outputs parameterized manipulation instructions. The robot controller calls the composite motion function, and the end effector is driven to execute the corresponding moxibustion manipulation trajectory according to the parameterized manipulation instructions.

[0043] For example, a 32-dimensional state feature vector is constructed, including: extracting the temperature distribution and temperature rise slope features of the acupoint area to obtain an 8-dimensional thermal field vector; extracting the a* channel component of skin redness and human displacement vector to obtain a 6-dimensional physiological displacement vector; obtaining the robot joint pose, linear velocity and target residual to obtain a 12-dimensional kinematic vector; collecting the three-dimensional contact force and three-dimensional torque received by the end effector to obtain a 6-dimensional interaction force feature vector; and concatenating the above vectors to obtain a 32-dimensional state feature vector.

[0044] This invention uses a multi-source feature fusion module to process perceptual data from different modalities into a unified 32-dimensional state feature vector, which is then used as input to a reinforcement learning model.

[0045] Specifically, the 32-dimensional feature vector is constructed as follows: Infrared thermal imaging features (8 dimensions): temperature at the center of the acupoint, average temperature around the acupoint, temperature gradient in the four quadrants, heating rate, and historical highest temperature shift. RGB visual physiological characteristics (6 dimensions): a* value of skin erythema component, area of ​​skin erythema region, displacement vector caused by human respiration. (3D), human movement frequency; Robotic arm pose and state characteristics (12-dimensional): Current position coordinates of the end effector in the base coordinate system (3D) Attitude angle of the end effector (3D), the current angular velocity or end-effector linear velocity of each joint. (3D), residual distance between the target point and the current point (3D); Torque sensor interaction characteristics (6-dimensional): Three-dimensional contact force on the end effector (3D) Three-dimensional torque on the end (3D).

[0046] Normalize the motion space As the output of the reinforcement learning model.

[0047] Mapped to physical parameters, as follows: Vertical velocity (Used to adjust the height of the heat source); Superimposed trajectory radius (Control the range of rotation); Superimposed trajectory frequency (Control the speed of the bird's pecking).

[0048] The moxibustion technique collaborative control method based on multi-source feature fusion provided in this invention can deeply characterize the thermal physiological response of the subject's skin and the micro-motion state of the human body by extracting a 32-dimensional feature vector that includes infrared, visual and force feedback.

[0049] In some embodiments of the present invention, the deep reinforcement learning policy network is trained with the objectives of maximizing cumulative reward and maximizing policy entropy; The reward function of the SAC algorithm is expressed as follows:

[0050] in, Represents the reward function, Indicates real-time temperature. Indicates the target treatment temperature. This indicates the rate of temperature change in the moxibustion area per unit time. A metric for evaluating the smoothness of a robot's motion trajectory. This indicates a positive feedback item for skin flushing. This indicates a penalty for crossing safety boundaries. , , and These represent the weighting coefficients.

[0051] Deep reinforcement learning policy networks employ the SAC algorithm, whose objective function includes maximizing cumulative reward and maximizing policy entropy. The reward function... Defined as:

[0052] in, Represents the reward function, For real-time temperature, Target treatment temperature, The reward is for motion smoothness (a negatively correlated term for acceleration). The rate of temperature change in the moxibustion area per unit time. This serves as an evaluation metric for the smoothness of a robot's motion trajectory. For skin flushing, positive feedback item. Penalties for crossing safety boundaries These are the weighting coefficients.

[0053] Figure 2 This is a schematic diagram of the deep reinforcement learning control network architecture provided by the present invention, as shown below. Figure 2 As shown, the output parameterized maneuver action instructions include: inputting a 32-dimensional state feature vector into a policy network based on a flexible actor-critic (SAC) framework; and sampling continuous action space parameters, including height adjustment speed, superimposed trajectory radius, superimposed frequency, and heat source power components, through the maximum entropy principle, as parameterized maneuver action instructions.

[0054] In some embodiments of the present invention, controlling the moxibustion robot to perform the corresponding moxibustion technique based on the action command includes: Based on the aforementioned action commands, the moxibustion robot is controlled to perform rotary moxibustion by superimposing a circular motion trajectory in the XY plane onto its basic pose; or, Based on the aforementioned action commands, the moxibustion robot is controlled to perform sparrow-pecking moxibustion by superimposing a reciprocating motion trajectory in the Z-axis direction on the basis of the basic posture.

[0055] The robot controller receives motion commands and uses a composite motion control interface to dynamically superimpose periodic geometric trajectories on the basic pose of the end effector, driving the actuator to complete the moxibustion action.

[0056] The end effector is driven by the parameterized technique action command to execute the corresponding moxibustion technique trajectory, including: controlling the robot controller to superimpose the circular motion trajectory of the XY plane on the reference pose to perform rotary moxibustion; or, superimposing the reciprocating motion trajectory of the Z axis on the reference pose to perform sparrow-pecking moxibustion.

[0057] Figure 3 This is a schematic diagram illustrating the principle of generating the sparrow-pecking moxibustion trajectory provided by the present invention. Figure 4 This is a schematic diagram illustrating the principle of generating the circular moxibustion trajectory provided by the present invention, as shown below. Figure 3 and Figure 4 As shown, the specific methods for implementing the technique using a composite motion control interface include: Sparrow-pecking moxibustion mode: The basic pose remains unchanged, and a geometric trajectory of "linear reciprocating" type is superimposed. Parameters include amplitude. H With frequency f ,in H,f Output in real time by the policy network.

[0058] Rotary Moxibustion Mode: The basic pose remains unchanged, and a geometric trajectory of type "planar circle" is superimposed. Parameters include the rotation radius. r With angular velocity ω ,in r , ω Output in real time by the policy network.

[0059] If the vibration frequency is within the preset frequency range and the signal energy is higher than the first preset energy threshold, the target moxibustion point is determined based on the current moxibustion point corresponding to the human body vibration signal.

[0060] Optionally, the moxibustion technique collaborative control method also includes a dynamic switching logic for the technique: in response to the temperature at the acupoint center rising to a preset temperature threshold and the skin redness component significantly increasing, a positive incentive function is triggered, the radius of rotation is controlled to decay linearly to zero, and reciprocating motion in the Z-axis direction is simultaneously activated to achieve a seamless switch from rotary moxibustion to sparrow-pecking moxibustion.

[0061] The moxibustion technique collaborative control method based on multi-source feature fusion provided in this invention solves the problem of stiff and non-continuous evolution of traditional moxibustion robot techniques by using parameterized technique adjustment through a deep reinforcement learning model, and achieves seamless switching and power coordination between techniques.

[0062] In some embodiments of the present invention, it further includes: When the temperature at the center of the acupoint exceeds the preset temperature or the positional deviation exceeds the preset displacement, the moxibustion robot is controlled to stop moxibustion and the robotic arm is driven to retract along the positive Z-axis.

[0063] This invention also includes a proactive safety intervention step, whereby the infrared thermal imaging detects a temperature exceeding an absolute safety threshold. When visual detection detects a sudden large displacement of the human body, the system triggers the highest priority shutdown interrupt through the hardware I / O interface and forces a positive Z-axis retraction action.

[0064] For example, during the execution of the moxibustion technique trajectory, the human body displacement vector and contact torque are monitored in real time. In response to the human body displacement vector exceeding a displacement threshold or the contact torque exceeding a torque threshold, the robot controller executes an emergency stop command and drives the end effector to retreat upwards a preset distance. Since the moxibustion process carries the risk of burns or collisions, real-time redundant monitoring using both visual and force sensors, combined with rapid retreat logic, can further enhance the safety of the system operation.

[0065] The moxibustion technique collaborative control method based on multi-source feature fusion provided in this invention ensures safety in complex human-computer interaction environments through multi-threshold safety monitoring and emergency retreat mechanisms.

[0066] Figure 5 This is a schematic diagram of the structure of the moxibustion robot provided by the present invention, as shown below. Figure 5 As shown, the present invention also provides a moxibustion robot, comprising: A six-axis robotic arm body, control unit, and end effector; The end effector is connected to the six-axis robotic arm body via a flange; The end-efficiency integrated actuator integrates an infrared thermal imager, an RGB camera, a six-dimensional torque sensor, and an electrically heated moxibustion head. The control device is used to execute the moxibustion technique collaborative control method based on multi-source feature fusion as described in any of the above implementations.

[0067] The moxibustion robot includes a six-axis robotic arm, an end effector, and a control device connected to the robotic arm and the end effector.

[0068] Figure 6 A schematic diagram of the end-effector integrated execution device provided by the present invention is shown below. Figure 6 As shown, the end-effector integrates an infrared thermal imager 601, an RGB camera 602, a six-dimensional torque sensor 603, and an electric heating moxibustion head 604.

[0069] The control device is used to execute the aforementioned collaborative control method for moxibustion techniques based on multi-source feature fusion and deep reinforcement learning, so as to realize intelligent moxibustion operation on the acupoint areas of the subject.

[0070] In some embodiments of the present invention, the infrared thermal imager, the RGB camera, the six-dimensional torque sensor, the electric heating moxibustion head, and the flange are coaxially mounted.

[0071] In some embodiments of the present invention, the infrared thermal imager is used to collect infrared thermal spectrum data of the moxibustion area; The RGB camera is used to acquire visible light image data of the moxibustion area; The six-dimensional torque sensor is used to collect contact torque data.

[0072] The infrared thermal imager, RGB camera, six-dimensional torque sensor, electric heating moxibustion head, and flange are coaxially mounted, with the six-dimensional torque sensor connected in series between the flange and the moxibustion head.

[0073] An infrared thermal imager is used to collect infrared thermal spectrum data of the moxibustion area, an RGB camera is used to collect visible light image data of the moxibustion area, and a six-dimensional torque sensor is used to collect contact torque data.

[0074] The moxibustion robot provided in this embodiment of the invention can realize the technical solution described in the above embodiment of the moxibustion technique collaborative control method based on multi-source feature fusion and deep reinforcement learning. The specific implementation principle can be found in the corresponding content in the above embodiment of the moxibustion technique collaborative control method based on multi-source feature fusion and deep reinforcement learning, and will not be repeated here.

[0075] Optionally, this embodiment also provides an integrated intelligent moxibustion system, which includes a six-axis collaborative robot, a main control computer, and an end effector. The main control computer communicates with the robot controller via Ethernet (TCP / IP protocol).

[0076] Figure 7 This is a schematic diagram of the overall process of the present invention, such as... Figure 7 As shown, the method for collaborative control of moxibustion techniques based on multi-source feature fusion and deep reinforcement learning provided by this invention includes: Step S701: Real-time acquisition of infrared thermal spectrum data, visible light images and contact torque data of the moxibustion area through a multi-source sensing module.

[0077] Step S702: Use spatiotemporal alignment and convolutional neural networks to extract multidimensional state features including acupoint temperature field gradient, skin redness features and human micro-motion vectors.

[0078] Step S703: Input the feature into a deep reinforcement learning model based on the flexible actor-critic (SAC) architecture, and output the spatial composite motion command and heat source power parameters of the robotic arm end effector.

[0079] Step S704: The control system calls the composite motion function to superimpose parameterized pecking or gyratory motion onto the basic trajectory.

[0080] Figure 8 A flowchart illustrating the collaborative control method provided by this invention is shown below. Figure 8 As shown, the collaborative control process is as follows: Step S801: Initialization and global positioning.

[0081] After the system starts, the robot controller executes a reset command, moving the end effector to directly above the target acupoint. The acupoint coordinates are locked via visual recognition, and the initial moxibustion height is set. At this point, the infrared sensor and torque sensor complete zero-point calibration, and the system enters real-time monitoring mode.

[0082] Step S802: Multimodal state perception and feature parsing.

[0083] The multi-source sensing module collects raw data in real time. A CNN network performs convolution processing on the infrared thermal image to extract the current acupoint center temperature. The temperature is 40℃. Simultaneously, the visual recognition module performs color component analysis on the skin area and determines that there are currently no redness features (redness component). (At the baseline threshold). At this point, the constructed 32-dimensional state vector This is characterized as the "insufficient preheating stage".

[0084] Step S803: Primary action decision based on SAC strategy.

[0085] The reinforcement learning agent receives the state vector. Based on the optimization objective of maximizing thermal permeability, the first-stage action command is output. .

[0086] This instruction includes a requirement to increase heat input, which is invoked through the controller using the composite motion interface: combine_motion_config(type=CIRCLE, ref_plane=XY, freq=0.5Hz, amp=20mm). At this point, the robotic arm maintains a constant Z-axis height, and the end effector performs a circular trajectory with a radius of 20mm in the XY plane, i.e., the rotary moxibustion mode.

[0087] Step S804: Execution of dynamic strategy adjustment and seamless switching of techniques driven by physiological feedback.

[0088] The system continuously monitors the multimodal physiological feedback data of the subject's skin during moxibustion. When the moxibustion duration reaches the preset observation period and the sensor detects the temperature at the center of the acupoint, the system will detect the temperature. The temperature rises to 43°C, and the visual module simultaneously detects the skin. When the redness component significantly increases, it is determined that the skin's thermal sensitivity response has been triggered. At this point, the positive activation term for redness in the deep reinforcement learning algorithm... Once activated, the agent determines that the current heat penetration depth has reached the target and immediately issues a method switching command, driving the robot controller to adjust the current planar rotation radius. Linear decay to 0. Without interrupting the basic motion of the robotic arm, the system seamlessly superimposes the variable frequency reciprocating motion parameters in the Z-axis direction, enabling the actuator to perform vertical reciprocating motion at a frequency of 1.5Hz and an amplitude of 30mm, thereby completing the dynamic mode switch from "rotary moxibustion" to "pecking moxibustion", and using high-frequency pulsed thermal radiation to further guide the heat to penetrate into the deep acupoint tissue.

[0089] Step S805, safety check.

[0090] Throughout the entire process, the system maintains the highest priority anomaly monitoring mechanism.

[0091] Step S806: Emergency avoidance and proactive safety intervention.

[0092] If the visual algorithm detects a sudden rise in the subject's back, resulting in a displacement vector If the preset safety threshold (e.g., 10mm) is exceeded, the system determines that there is a risk of collision or burn, immediately interrupts the current motion sequence, calls the stop_motion() instruction, and drives the robotic arm to quickly retract 100mm along the positive Z-axis.

[0093] The moxibustion technique collaborative control method provided in this invention extracts a 32-dimensional feature vector including infrared, visual, and force feedback data through a multi-source sensing processing module, which can deeply characterize the subject's skin's thermosensitive physiological response and the body's micro-motion state. The parameterized technique adjustment achieved using a deep reinforcement learning model solves the problems of stiff and uncontinuous evolution in traditional moxibustion robot techniques, realizing seamless switching and power coordination between techniques. Furthermore, multi-threshold safety monitoring and an emergency retreat mechanism ensure safety in complex human-computer interaction environments. This application can automatically optimize the moxibustion trajectory and heat based on real-time user feedback without human intervention, solving the problem that moxibustion results are greatly affected by the operator's experience, and significantly improving the biomimetic effect and user experience of moxibustion.

[0094] like Figure 9 As shown, the present invention also provides an electronic device 900. The electronic device 900 includes a processor 901, a memory 902, and a display 903. Figure 9 Only some components of the electronic device 900 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0095] In some embodiments, processor 901 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 902 or process data, such as the moxibustion technique collaborative control method based on multi-source feature fusion in this invention.

[0096] In some embodiments, processor 901 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 901 may be local or remote. In some embodiments, processor 901 may be implemented on a cloud platform. In some embodiments, the cloud platform may include private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, etc., or any combination thereof.

[0097] In some embodiments, memory 902 may be an internal storage unit of electronic device 900, such as a hard disk or memory of electronic device 900. In other embodiments, memory 902 may also be an external storage device of electronic device 900, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 900.

[0098] Furthermore, the memory 902 may include both internal storage units of the electronic device 900 and external storage devices. The memory 902 is used to store application software and various types of data installed on the electronic device 900.

[0099] In some embodiments, display 903 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an organic light-emitting diode (OLED) touchscreen. Display 903 is used to display information from electronic device 900 and to display a visual user interface. Components 901-903 of electronic device 900 communicate with each other via a system bus.

[0100] In one embodiment, when the processor 901 executes the moxibustion technique collaborative control program based on multi-source feature fusion in the memory 902, the following steps can be implemented: Acquire target data of the moxibustion area and extract target features from the target data to construct a state vector; the target data includes infrared thermal spectrum data, visible light image data, and contact torque data of the end effector of the moxibustion robot arm; the target features include acupoint center temperature features, skin redness features, human body displacement features, and torque features; The state vector is input into the trained deep reinforcement learning policy network to obtain the action command output by the deep reinforcement learning policy network; the deep reinforcement learning policy network adopts the SAC algorithm; the action command includes the spatial velocity vector of the robotic arm end effector, trajectory superposition parameters, and heating power of the heat source; Based on the action commands, the moxibustion robot is controlled to perform the corresponding moxibustion techniques.

[0101] It should be understood that when the processor 901 executes the moxibustion technique collaborative control program based on multi-source feature fusion in the memory 902, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0102] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 900 mentioned. Electronic device 900 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 900 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0103] Accordingly, embodiments of the present invention also provide a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the moxibustion technique collaborative control method based on multi-source feature fusion provided in the above-described method embodiments.

[0104] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0105] The above provides a detailed description of the moxibustion technique collaborative control method and moxibustion robot based on multi-source feature fusion provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for collaborative control of moxibustion techniques based on multi-source feature fusion, characterized in that, include: Acquire target data of the moxibustion area and extract target features from the target data to construct a state vector; The target data includes infrared thermal spectrum data, visible light image data, and contact torque data at the end of the moxibustion robot's robotic arm; the target features include acupoint center temperature features, skin redness features, human body displacement features, and torque features. The state vector is input into the trained deep reinforcement learning policy network to obtain the action command output by the deep reinforcement learning policy network; the deep reinforcement learning policy network adopts the SAC algorithm; the action command includes the spatial velocity vector of the robotic arm end effector, trajectory superposition parameters, and heating power of the heat source; Based on the action commands, the moxibustion robot is controlled to perform the corresponding moxibustion techniques.

2. The method for coordinated control of moxibustion techniques based on multi-source feature fusion according to claim 1, characterized in that, The deep reinforcement learning policy network is trained with the objectives of maximizing cumulative reward and maximizing policy entropy. The reward function of the SAC algorithm is expressed as follows: in, Represents the reward function, Indicates real-time temperature. Indicates the target treatment temperature. This indicates the rate of temperature change in the moxibustion area per unit time. A metric for evaluating the smoothness of a robot's motion trajectory. This indicates a positive feedback item for skin flushing. This indicates a penalty for crossing safety boundaries. , , and These represent the weighting coefficients.

3. The method for coordinated control of moxibustion techniques based on multi-source feature fusion according to claim 1, characterized in that, The step of controlling the moxibustion robot to perform corresponding moxibustion techniques based on the action commands includes: Based on the aforementioned action commands, the moxibustion robot is controlled to perform rotary moxibustion by superimposing a circular motion trajectory in the XY plane onto its basic pose; or, Based on the aforementioned action commands, the moxibustion robot is controlled to perform sparrow-pecking moxibustion by superimposing a reciprocating motion trajectory in the Z-axis direction on the basis of the basic posture.

4. The method for coordinated control of moxibustion techniques based on multi-source feature fusion according to claim 1, characterized in that, The expression for the state vector is as follows: in, Represents the state vector. Indicates the temperature at the center of the acupoint. This represents the temperature field distribution gradient centered on the acupoint. This represents the skin redness component in the Lab color space. This indicates positional deviation caused by slight movements of the human body. Indicates the end contact torque. This indicates the remaining duration of moxibustion treatment.

5. The method for coordinated control of moxibustion techniques based on multi-source feature fusion according to claim 4, characterized in that, Also includes: When the temperature at the center of the acupoint exceeds the preset temperature or the positional deviation exceeds the preset displacement, the moxibustion robot is controlled to stop moxibustion and the robotic arm is driven to retract along the positive Z-axis.

6. A moxibustion robot, characterized in that, include: A six-axis robotic arm body, control unit, and end effector; The end effector is connected to the six-axis robotic arm body via a flange; The end-efficiency integrated actuator integrates an infrared thermal imager, an RGB camera, a six-dimensional torque sensor, and an electrically heated moxibustion head. The control device is used to execute the moxibustion technique collaborative control method based on multi-source feature fusion as described in any one of claims 1 to 5.

7. The moxibustion robot according to claim 6, characterized in that, The infrared thermal imager, the RGB camera, the six-dimensional torque sensor, the electric heating moxibustion head, and the flange are coaxially mounted.

8. The moxibustion robot according to claim 6, characterized in that, The infrared thermal imager is used to collect infrared thermal spectrum data of the moxibustion area; The RGB camera is used to acquire visible light image data of the moxibustion area; The six-dimensional torque sensor is used to collect contact torque data.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the moxibustion technique collaborative control method based on multi-source feature fusion as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the moxibustion technique collaborative control method based on multi-source feature fusion as described in any one of claims 1 to 5.

Images